MAS4DT: A proposal for developing Digital Twins based-on Multi-Agent Systems

Digital Twins (DTs) have emerged as a transformative technology across sectors such as manufacturing, smart cities, healthcare, and education, enabling real-time monitoring and control of physical assets. Nevertheless, their practical application continues to face significant challenges. Among the most significant are the need to establish design and modelling approaches capable of faithfully representing the essential attributes of physical assets; the ability to ensure precise control of their physical counterparts; the requirement to sustain continuous and reliable synchronisation in real-time; and the challenge of composing multiple DTs to support the management of complex systems. These difficulties are compounded by the current lack of standardised frameworks and methodologies for DT development, as well as the increasing complexity of applications that demand more than digital models or shadows, seeking not only to monitor the physical asset but also to interact with and control it in an intelligent and autonomous manner. To address these issues, this thesis explores the use of the Multi-Agent System (MAS) paradigm as a foundational approach for supporting the design, modelling, and development of DTs. A MAS consists of a collection of autonomous, interacting agents that collaborate or compete to achieve individual or collective goals within a shared environment. MASs offer a natural solution for orchestrating distributed, autonomous, and intelligent behaviours that enable DTs that are not only reactive but also proactive, scalable, and capable of complex coordination and decision-making. Firstly, the thesis begins with the systematic literature reviews on the integration between Digital Twins (DT) and Multi-Agent Systems (MAS). These reviews identifies the main existing synergies between the properties of both paradigms, as well as the research gaps and the specific challenges associated with Human Digital Twins (HDTs). Secondly, the thesis develops MAS4DT, a novel domain-independent methodology based on MAS principles. This methodology guides stakeholders through the systematic design of DTs using a structured and iterative process. MAS4DT was empirically validated through a controlled experiment, designed according to experimental guidelines in software engineering. Compared to other DT design methodologies, MAS4DT is superior in terms of understandability. Thirdly, the thesis focuses on the advanced design of three properties of DT: entanglement, predictability, and composability. For the entanglement property, it proposes a catalogue of agent design patterns that ensure a close connection between the DT and its physical counterpart. Regarding the predictability property, it applies the artificial intelligence technique of active inference, enabling DTs to simulate the future behaviour of the physical twin and adapt proactively. With regard to the composability property, the thesis analyses the role of artificial collective intelligence as a mechanism to facilitate the transition from traditional DTs to autonomous DTs (ADTs), which are capable of operating collaboratively within a federated DT (FDT). The results confirm that MAS properties such as autonomy, proactivity, and sociability enhance the capabilities of DTs. Furthermore, the MAS4DT methodology, together with the proposed patterns and guidelines, facilitates the creation of more intelligent DTs. This thesis provides a structured approach to exploiting MAS in the development of DT, offering theoretical knowledge and practical tools for researchers and practitioners, and laying a solid foundation for future work on autonomous and federated DTs.